Large Vision-Language Model Security: A Survey

Taowen Wang; Zheng Fang; Haochen Xue; Chong Zhang; Mingyu Jin; Wujiang Xu; Dong Shu; Shanchieh  Yang; Zhenting Wang; Dongfang Liu

Large Vision-Language Model Security: A Survey

Taowen Wang, Zheng Fang, Haochen Xue, Chong Zhang, Mingyu Jin, Wujiang Xu, Dong Shu, Shanchieh Yang, Zhenting Wang, Dongfang Liu^*

^*Corresponding author for this work

Research output: Contribution to journal › Conference article › peer-review

61 Downloads (Pure)

Abstract

In the domain of Large Vision-Language Models (LVLMs), securing these models has emerged as a critical issue for both researchers and practitioners. In this paper, we highlight and analyze the security-related issues of LVLMs, with a special emphasis on the reliability challenges in practical deployments. We begin by reviewing recent studies on threats like jailbreak and backdoor attacks, alongside discussing the potential countermeasures implemented to mitigate these risks. Additionally, we touch on real-world application problems, such as hallucinations and privacy leakages, as well as the ethical and legal related researches around them. We also outline the shortcomings observed in current studies and discuss directions for future research, with the aim of promoting LVLMs towards a safer direction. A curated list of LVLMs-security-related resources is also available at https://github.com/MingyuJ666/LVLM-Safety.

Original language	English
Article number	1
Pages (from-to)	3
Number of pages	22
Journal	Communications in Computer and Information Science
Volume	2315
Publication status	Published - 27 Dec 2024

Access to Document

Large Vision-Language Model Security: A SurveyFinal published version, 942 KBLicence: CC BY

Cite this

@article{e9676ed327c649e3bd4fe7ddee421c63,

title = "Large Vision-Language Model Security: A Survey",

abstract = "In the domain of Large Vision-Language Models (LVLMs), securing these models has emerged as a critical issue for both researchers and practitioners. In this paper, we highlight and analyze the security-related issues of LVLMs, with a special emphasis on the reliability challenges in practical deployments. We begin by reviewing recent studies on threats like jailbreak and backdoor attacks, alongside discussing the potential countermeasures implemented to mitigate these risks. Additionally, we touch on real-world application problems, such as hallucinations and privacy leakages, as well as the ethical and legal related researches around them. We also outline the shortcomings observed in current studies and discuss directions for future research, with the aim of promoting LVLMs towards a safer direction. A curated list of LVLMs-security-related resources is also available at https://github.com/MingyuJ666/LVLM-Safety.",

author = "Taowen Wang and Zheng Fang and Haochen Xue and Chong Zhang and Mingyu Jin and Wujiang Xu and Dong Shu and Shanchieh Yang and Zhenting Wang and Dongfang Liu",

year = "2024",

month = dec,

day = "27",

language = "English",

volume = "2315",

pages = "3",

journal = "Communications in Computer and Information Science",

issn = "1865-0929",

}

TY - JOUR

T1 - Large Vision-Language Model Security: A Survey

AU - Wang, Taowen

AU - Fang, Zheng

AU - Xue, Haochen

AU - Zhang, Chong

AU - Jin, Mingyu

AU - Xu, Wujiang

AU - Shu, Dong

AU - Yang, Shanchieh

AU - Wang, Zhenting

AU - Liu, Dongfang

PY - 2024/12/27

Y1 - 2024/12/27

N2 - In the domain of Large Vision-Language Models (LVLMs), securing these models has emerged as a critical issue for both researchers and practitioners. In this paper, we highlight and analyze the security-related issues of LVLMs, with a special emphasis on the reliability challenges in practical deployments. We begin by reviewing recent studies on threats like jailbreak and backdoor attacks, alongside discussing the potential countermeasures implemented to mitigate these risks. Additionally, we touch on real-world application problems, such as hallucinations and privacy leakages, as well as the ethical and legal related researches around them. We also outline the shortcomings observed in current studies and discuss directions for future research, with the aim of promoting LVLMs towards a safer direction. A curated list of LVLMs-security-related resources is also available at https://github.com/MingyuJ666/LVLM-Safety.

AB - In the domain of Large Vision-Language Models (LVLMs), securing these models has emerged as a critical issue for both researchers and practitioners. In this paper, we highlight and analyze the security-related issues of LVLMs, with a special emphasis on the reliability challenges in practical deployments. We begin by reviewing recent studies on threats like jailbreak and backdoor attacks, alongside discussing the potential countermeasures implemented to mitigate these risks. Additionally, we touch on real-world application problems, such as hallucinations and privacy leakages, as well as the ethical and legal related researches around them. We also outline the shortcomings observed in current studies and discuss directions for future research, with the aim of promoting LVLMs towards a safer direction. A curated list of LVLMs-security-related resources is also available at https://github.com/MingyuJ666/LVLM-Safety.

UR - https://link.springer.com/chapter/10.1007/978-981-96-0151-6_1

M3 - Conference article

SN - 1865-0929

VL - 2315

SP - 3

JO - Communications in Computer and Information Science

JF - Communications in Computer and Information Science

M1 - 1

ER -

Large Vision-Language Model Security: A Survey

Abstract

Access to Document

Other files and links

Fingerprint

Cite this